A new energy load growth prediction combination-based energy storage configuration method

Abstract

The application discloses a kind of energy storage configuration methods based on source load growth prediction combination, comprising:1, extraction daily load curve, conventional unit and new energy daily output curve, collect the date type in each sampling period;2, respectively using k-means algorithm to the sampling object clustering according to date type;3, respectively using LSTM algorithm to the sampling object prediction, respectively form daily load, conventional unit and new energy daily output clustering prediction result;4, the result obtained in step 3 is combined, eliminate the clustering prediction result under the same clustering factor;5, calculate the net load fluctuation corresponding to different combinations, build the energy storage configuration model about the uncertainty of source load two sides;6, the model is solved using particle swarm optimization algorithm.The application can make source load side grow under the premise of medium and long term operation, configure energy storage capacity, so that power grid can maintain power balance under medium and long term operation period.

Description

Technical Field

[0001] This invention belongs to the field of energy storage configuration technology, specifically relating to an energy storage configuration method based on a combination of new energy load growth forecasts. Background Technology

[0002] As the grid load gradually increases, the power supply pressure on the grid supply side gradually increases. In recent years, in order to alleviate the grid pressure, it has been necessary to adopt peak shaving and valley filling measures to balance the power deviation between the generation side and the load side of the power system.

[0003] With the introduction of the dual-carbon development plan, my country is vigorously developing new energy sources. However, new energy sources have significant negative impacts due to their volatility and randomness. Therefore, existing research generally treats either the power generation side or the load side of new energy sources separately, based on peak shaving and valley filling. However, to accelerate the construction of new energy sources and reduce pollution from traditional energy power generation, the installed capacity of new energy sources is also increasing year by year. Therefore, treating either the power generation side or the load side of new energy sources separately cannot accurately predict the changes in grid power in the medium and long term.

[0004] Energy storage is a crucial means of achieving dual-carbon goals. It is simple to control and can effectively address power imbalances at both ends of the power grid. Therefore, failing to simultaneously consider power changes on both the generation and load sides when predicting grid power variations can significantly impact energy storage deployment planning. In conclusion, existing energy storage deployment plans do not account for changes in power on both sides, making it difficult to meet the requirements of energy storage deployment planning. Summary of the Invention

[0005] The present invention aims to address the shortcomings of the existing technology by proposing an energy storage configuration method based on the prediction of new energy load growth. This method aims to simultaneously consider the growth changes on both the load side and the new energy output side, enabling the planned energy storage device to be used in the grid for a long period of time.

[0006] To achieve the above-mentioned objectives, the present invention adopts the following technical solution:

[0007] The present invention provides an energy storage configuration method based on source-load growth prediction combination, characterized by the following steps:

[0008] Step 1: Extract historical data, including: daily load curves, daily power output curves of conventional units and new energy units, and the date type of each curve, including: seasonal type, weekday and holiday;

[0009] Step 2: Collect sample data from each curve, and use the k-means algorithm to cluster the daily load sample data under different date types, obtaining the daily load clustering results under different date types; cluster the daily power output sample data of conventional units and the daily power output sample data of new energy units according to the season type, obtaining the daily power output clustering results of conventional units and the daily power output clustering results of new energy units under different season types.

[0010] Step 3: Input the three clustering results into the LSTM model respectively to predict the daily load and the daily power output on the power supply side, and obtain the corresponding daily load clustering prediction result P. L Clustering prediction results of daily power output of conventional generating units P G Clustering prediction results of daily power output of new energy units P New ;

[0011] Step 4: Predict the daily load from the clustering results P of the p types of daily loads L Clustering prediction results of daily power output of g-type conventional generating units P G Clustering prediction results of q kinds of new energy sources P New The r cluster prediction results under different date types are removed, and then s = p × g × qr daily load corresponding to the daily power generation cluster prediction results on the power supply side are combined.

[0012] Step 5: Calculate the net load fluctuation corresponding to the daily power generation clustering prediction results on the power supply side for each type of daily load, and consider the uncertainty of the daily power generation of new energy sources on the power supply side and the uncertainty on the load side to construct an energy storage planning model for the uncertainty on both the source and load sides.

[0013] Step 6: Use the particle swarm optimization algorithm to solve the energy storage configuration model with respect to the uncertainties on both sides of the source and load, so as to obtain the optimal energy storage configuration scheme.

[0014] The energy storage configuration method based on source-load growth prediction combination described in this invention is also characterized in that step 5 includes:

[0015] Step 5.1: Calculate the net load fluctuation corresponding to the daily power generation clustering prediction result on the power supply side under the i-th clustering prediction combination using equation (1):

[0016] ΔP i =P L,i -P G,i -P New,i (1)

[0017] In equation (1), P L,i P G,i P New,iThe load clustering prediction results, conventional unit output clustering prediction results, and new energy output clustering prediction results are corresponding to the i-th clustering prediction combination results obtained in step 3.

[0018] Step 5.2: Use equation (2) to obtain the net load power uncertainty r of the daily power generation cluster prediction results corresponding to the power source side under the i-th cluster prediction combination. NL,i :

[0019]

[0020] In equation (3), r L,i Let r be the load growth uncertainty under the i-th clustering prediction combination. New,i The uncertainty of new energy output under the i-th cluster prediction combination;

[0021] Using equation (3), the net load fluctuation P considering uncertainty is obtained. rmin P rmax And enable the maximum energy storage configuration capacity E E,max In [P] rmin ,P rmax ]Inside;

[0022]

[0023] Step 5.3: With the goal of minimizing load fluctuations, construct the objective function of the energy storage configuration model regarding the uncertainty of load and new energy output using equation (4):

[0024] min C total =C ch +C sys (4)

[0025] In equation (4): C total C represents the total configuration cost. ch C represents the total cost of energy storage, which is obtained from equation (5); sys The system operating cost is given by equation (6);

[0026]

[0027]

[0028] In equations (5) and (6): C C Indicates the lifespan cost of energy storage; C inv For energy storage construction costs; C ope For energy storage operation and maintenance costs; μ is the discount rate; κ is the energy storage lifespan; C E Cost of energy storage battery capacity; C P Cost per unit power; E E P represents the total configured capacity of energy storage.E C represents the total configured power of energy storage. o Daily operating and maintenance cost per unit; P c The daily charge / discharge power of the energy storage; T is the prediction period; c G C is the operating cost of thermal power units; on C off For the start-up and shutdown costs of thermal power units; c voll Cost per unit load shedding; To predict the system load shedding at time t within the prediction period; c New The cost of energy curtailment for new energy units; This is to predict the energy discarded by new energy sources at time t within the predicted period.

[0029] The total energy storage cost C in the energy storage configuration model ch =C C +C inv +C ope Among them, C C The energy storage lifetime cost is represented by equation (7);

[0030]

[0031] In equation (7), N represents the total number of energy storage cycles within the planning period; N eq,n For the nth equivalent cycle of energy storage, and obtained from equation (8), c DOD The cost of performing a deep charge-discharge cycle on the battery is obtained from equation (9):

[0032]

[0033]

[0034] In equations (8) and (9): N eq N is the equivalent coefficient; en The number of battery cycles at 100% depth of discharge; c DOD Cost per cycle at 100% depth of discharge; D DOD This represents the actual depth of cycle discharge, and D DOD = SOC1 - SOC2, where SOC1 represents the initial charge / discharge capacity of the energy storage battery, and SOC2 represents the final charge / discharge capacity of the energy storage battery; N roll This refers to the number of battery cycles at 100% depth of discharge.

[0035] The constraints of the energy storage planning model are constructed using equations (11)-(19):

[0036] P G,t +P New,t +βP ch,t =PL,t (11)

[0037] P ij,t -b ij (θ i,t -θ j,t )=0 (12)

[0038] 0≤P G,t ≤P G,max (13)

[0039] 0≤P New,t ≤P New,max (14)

[0040]

[0041]

[0042] SOC min ≤SOC t ≤SOC max (17)

[0043] P E ≤P E,max (18)

[0044] E E ≤E E,max (19)

[0045] In equations (11)-(19): P New,t P C,t P L,t Let be the new energy output, energy storage output, and load of the system at time t, respectively, and β be the energy storage operating state. Where X in In the energy storage charging state, X out In the energy storage and discharge state; P ij,t To predict the power between nodes i and j at time t, b ij Let θ be the susceptance within nodes i and j. i,t θ j,t These represent the phase angles of nodes i and j at time t, respectively; SOC t To predict the state of charge at time t, SOC min SOC max These represent the maximum and minimum permissible depth of discharge for the battery, respectively; E max This represents the maximum permissible capacity of the energy storage system.

[0046] The present invention provides an electronic device, including a memory and a processor, wherein the memory is used to store a program that supports the processor in executing the energy storage configuration method, and the processor is configured to execute the program stored in the memory.

[0047] The present invention discloses a computer-readable storage medium on which a computer program is stored, wherein the computer program is executed by a processor to perform the steps of the energy storage configuration method.

[0048] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0049] 1. This invention combines clustering and prediction algorithms. Since traditional prediction algorithms only consider time series, while clustering algorithms generally consider the characteristic features of clustered objects, this invention combines the advantages of both algorithms. For the clustered and predicted objects, it can explore the changing patterns of objects in terms of characteristics and time series, effectively improving the prediction accuracy.

[0050] 2. This invention simultaneously calculates the cluster prediction results of load and renewable energy output. Considering that changes in both the load side and renewable energy output side will affect the power of the power grid, the invention calculates the net load fluctuation under different results and selects the maximum value as the required energy storage capacity, thereby reducing the possibility of grid instability caused by power imbalance on both sides. Attached Figure Description

[0051] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0052] In this embodiment, as Figure 1 As shown, an energy storage configuration planning method based on a combination of new energy load growth forecasts includes the following steps:

[0053] Step 1: Extract historical data, which includes daily load curves, conventional unit output curves, and renewable energy daily output curves, as well as the date type for the corresponding dates. Date types include seasonal type, weekday, and holiday.

[0054] Step 2: Collect sample data from each curve, and use the k-means algorithm to cluster the daily load sample data under different date types, obtaining the daily load clustering results under different date types; cluster the daily power output sample data of conventional units and the daily power output sample data of new energy units according to the season type, obtaining the daily power output clustering results of conventional units and the daily power output clustering results of new energy units under different season types.

[0055] Daily loads are categorized according to the date type mentioned in step 1, with an initial range of 4 to 6. Conventional and renewable energy daily power generation are categorized according to season type, with an initial range of 3 to 4 for both categories. Based on the daily load, conventional unit, and renewable energy daily power generation data obtained in step 1, the number of categories k is selected (initially k1 = 4, k2 = 3, k3 = 3). k1 daily load curves are randomly selected from the daily load sample data, k2 conventional unit daily power generation curves are selected from the conventional unit daily power generation sample data, and k3 renewable energy daily power generation curves are selected from the renewable energy daily power generation sample data. These are used as the initial cluster centers for daily load, conventional unit daily power generation, and renewable energy daily power generation, respectively.

[0056] Using equation (1), calculate the Euclidean distance between the remaining sample data curves and the initial cluster center, and classify the data curve with the shortest Euclidean distance and its corresponding cluster center into the same category:

[0057]

[0058] In equation (1): dis is the Euclidean distance between curves; This is the m-th curve outside the initial cluster centers; Let n be the initial cluster center and n be the nth curve, where n = 1 to k.

[0059] Calculate the average value within the same category and use it as the new cluster center. Repeat steps 2.3 and 2.4 until the cluster centers no longer change significantly, and output the classification results for category k (including k1, k2, and k3).

[0060] Step 3: Using the clustering results obtained in Step 2 as input data, the LSTM algorithm is used to predict the daily load and the daily power generation on the power source side, respectively, to form the clustering prediction results of the daily load and the daily power generation of new energy sources.

[0061] The daily load clustering results, conventional unit output clustering results, and new energy output clustering results obtained in step 2 are used as data subsets, and the data under different classification results are sorted by date. Under the same classification result, the first 80% of the data in the clustering results are used as the training set, and the last 20% of the data are used as the prediction set to train the LSTM model. The specific training process of the LSTM model is shown in equations (2)-(7);

[0062] f t =σ(W f ·[h t-1 ,x t ]+b f (2)

[0063] i t =σ(Wi ·[h t-1 ,x t ]+b i (3)

[0064]

[0065]

[0066] o t =σ(W i ·[h t-1 ,x t ]+b o (6)

[0067] h t =o t *tanh C t (7)

[0068] In the formula: W f W i W o W C Please provide the weight coefficients for the forget gate, input gate, output gate, and state unit, respectively. f b i b C b o σ is the offset constant, and σ is the sigmoid function.

[0069] Based on the trained LSTM prediction model, the daily load, daily power output of conventional units, and daily power output of renewable energy are input to predict the daily load, daily power output of conventional units, and daily power output of renewable energy in future time segments.

[0070] Step 4: Let the daily load clustering prediction result P L There are p types, and the clustering prediction result of the conventional unit output is P. G There are g types, and the clustering prediction result P for new energy output is... New There are q kinds of results. After eliminating r kinds of results that are not in the same clustering factor, we can combine them to get p×g×qr kinds of daily load corresponding to the daily power generation results on the power supply side.

[0071] After removing clustering prediction results from different date types, in step 2, daily load is categorized by date type, and conventional unit output and renewable energy output are categorized by season type. After obtaining the clustering results, step 3 is performed to obtain the clustered prediction results. The daily load clustering prediction results, conventional unit output clustering prediction results, and renewable energy output clustering prediction results under the same season type are combined. Since the resulting clustering results are all seasonally related, they are processed after combination. For example, clustering results highly correlated with winter and clustering results highly correlated with summer cannot be combined and must be removed.

[0072] Step 5: Calculate the net load fluctuation corresponding to the different combination structures obtained in Step 4, and construct an energy storage planning model for uncertainties on both the power source and load sides, taking into account the uncertainties of new energy output on the power source side and the load side.

[0073] Using equation (8), calculate the net load fluctuation corresponding to the daily power generation clustering prediction result on the power supply side under the i-th clustering prediction combination:

[0074] ΔP i =P L,i -P G,i -P New,i (8)

[0075] In equation (8), P L,i P G,i P New,i The load clustering prediction results, conventional unit output clustering prediction results, and renewable energy output clustering prediction results corresponding to the i-th clustering prediction combination obtained in step 3 are:

[0076] Using equation (9), the net load power uncertainty r of the daily power generation cluster prediction results corresponding to the daily load on the power source side under the i-th cluster prediction combination is obtained. NL,i :

[0077]

[0078] In equation (9), r L,i Let r be the load growth uncertainty under the i-th clustering prediction combination. New,i The uncertainty of new energy output under the i-th cluster prediction combination;

[0079] Using equation (10), the net load fluctuation P considering uncertainty is obtained. rmin P rmax And enable the maximum energy storage configuration capacity E E,max In [P] rmin ,P rmax ]Inside;

[0080]

[0081] An energy storage planning model is constructed to address the uncertainties in load and renewable energy output. The objective function is shown in equation (11):

[0082] min C total =C ch +C sys (11)

[0083]

[0084]

[0085] In the formula: C total C represents the total planning cost. ch C represents the total cost of energy storage. sys For system operating costs; C inv For energy storage construction costs; C ope For energy storage operation and maintenance costs; μ is the discount rate; κ is the energy storage lifespan; C E Cost of energy storage battery capacity; C P Cost per unit power; E E P represents the total configured capacity of energy storage. E C represents the total configured power of energy storage. o Daily operating and maintenance cost per unit; P c c is the daily charge and discharge power of energy storage; G C is the operating cost of thermal power units; on C off For the start-up and shutdown costs of thermal power units; c voll Cost per unit load shedding; To predict the system load shedding at time t; c New The cost of energy curtailment for new energy units; To predict the energy discarded by new energy sources at time t.

[0086] The total energy storage cost C in the energy storage configuration model based on source-load growth forecasting combination ch =C C +C inv +C ope Among them, C C The energy storage lifetime cost is represented by equation (14);

[0087]

[0088] In equation (14), N represents the total number of energy storage cycles within the planning period; N eq,n For the nth equivalent cycle of energy storage, and obtained from equation (15), c DOD The cost of performing a deep charge-discharge cycle on the battery is given by equation (16):

[0089]

[0090]

[0091] In equations (15) and (16): N eq N is the equivalent coefficient; en The number of battery cycles at 100% depth of discharge; c DOD Cost per cycle at 100% depth of discharge. (D) DOD This represents the actual depth of cycle discharge, and D DOD = SOC1 - SOC2, where SOC1 represents the initial charge / discharge capacity of the energy storage battery, and SOC2 represents the final charge / discharge capacity of the energy storage battery; N roll This refers to the number of battery cycles at 100% depth of discharge.

[0092] Therefore, the total cost of energy storage should include C. C C ch =C C +C inv +C ope .

[0093] The constraints are as follows:

[0094] P G,t +P New,t +βP ch,t =P L,t (17)

[0095] P ij,t -b ij (θ i,t -θ j,t )=0 (18)

[0096] 0≤P G,t ≤P G,max (19)

[0097] 0≤P New,t ≤P New,max (20)

[0098]

[0099]

[0100] SOC min ≤SOC t ≤SOC max (twenty three)

[0101] P E ≤P E,max (twenty four)

[0102] E E ≤E E,max (25)

[0103] In the formula: Equation (17) is the power balance equation, P New,t P C,t P L,t Let be the renewable energy output, energy storage output, and load of the system at time t, respectively, and β be the energy storage operating state. Where X in In the energy storage charging state, X out This represents the energy storage discharge state; equation (18) represents the power flow constraint, P ij,t Let b be the power between nodes i and j at time t. ij Let θ be the susceptance within nodes i and j. i,t θ j,t Let i and j be the phase angles at time t, respectively; Equations (19)-(20) are the output constraints of conventional units and new energy sources; Equations (21)-(22) are the load shedding constraints and new energy curtailment constraints; Equation (23) is the state of charge constraint of energy storage, SOC. t Let SOC be the state of charge at time t. min SOC max Equations (24) and (25) represent the maximum and minimum allowable depths of discharge for the battery, respectively; Equations (24) and (25) represent the power and capacity constraints for the energy storage configuration. max This represents the maximum permissible capacity of the energy storage system.

[0104] Step 6: Use the particle swarm optimization algorithm to solve the energy storage planning model with respect to uncertainties on both the source and load sides, thereby obtaining the optimal energy storage configuration scheme. The specific steps are as follows:

[0105] Initialize particle swarm parameters, including race variables and iteration count; update the population using an adaptive particle swarm algorithm based on the objective function and constraints; adjust the solution speed and inertia factor in each iteration to avoid getting trapped in local solutions and update the iteration count; terminate the process and output the required energy storage capacity when the maximum number of iterations is reached; otherwise, return to continue optimizing the solution.

[0106] In this embodiment, an electronic device includes a memory and a processor. The memory stores a program that supports the processor in executing the above-described method, and the processor is configured to execute the program stored in the memory.

[0107] In this embodiment, a computer-readable storage medium stores a computer program, which is executed by a processor to perform the steps of the above method.

Claims

1. A method for configuring energy storage based on source-load growth prediction, characterized in that, Includes the following steps: Step 1: Extract historical data, including: daily load curves, daily power output curves of conventional units and new energy units, and the date type of each curve, including: seasonal type, weekday and holiday; Step 2: Collect sample data from each curve, and use the k-means algorithm to cluster the daily load sample data under different date types, obtaining the daily load clustering results under different date types; cluster the daily power output sample data of conventional units and the daily power output sample data of new energy units according to the season type, obtaining the daily power output clustering results of conventional units and the daily power output clustering results of new energy units under different season types. Step 3: Input the three clustering results into the LSTM model respectively to predict the daily load and the daily power output on the power supply side, and obtain the corresponding daily load clustering prediction results. P L Clustering prediction results of daily power output of conventional generating units P G Clustering prediction results of daily power output of new energy units P New ; Step 4: From p Daily load clustering prediction results P L , g Clustering prediction results of daily power output of conventional generating units P G , q Clustering prediction results of new energy sources P New Remove dates from different date types r Clustering prediction results are combined to form s = p × g × q - r Clustering prediction results of daily power output on the power supply side corresponding to different daily loads; Step 5: Calculate the net load fluctuation corresponding to the daily power generation clustering prediction results on the power supply side for each type of daily load, and consider the uncertainty of the daily power generation of new energy sources on the power supply side and the uncertainty on the load side to construct an energy storage planning model for the uncertainty on both the source and load sides. Step 6: Use the particle swarm optimization algorithm to solve the energy storage configuration model with respect to the uncertainties on both sides of the source and load, so as to obtain the optimal energy storage configuration scheme.

2. The energy storage configuration method based on source-load growth prediction combination according to claim 1, characterized in that, Step 5 includes: Step 5.1: Calculate the first step using equation (1). i The net load fluctuation corresponding to the daily power generation clustering prediction results on the power supply side under the following clustering prediction combinations: (1) In equation (1), P L,i , P G,i , P New,i The first one obtained in step 3 i The load clustering prediction results, conventional unit output clustering prediction results, and new energy output clustering prediction results corresponding to the clustering prediction combination results; Step 5.2: Use equation (2) to obtain the first... i Uncertainty of net load power in the daily power generation clustering prediction results corresponding to the daily load under the clustering prediction combination. r NL,i : (2) In equation (3), r L,i For the first i Uncertainty of load growth under clustering prediction combination r New,i For the first i Uncertainty in new energy output under clustering prediction combinations; Equation (3) is used to obtain the net load fluctuation considering uncertainty. P rmin , P rmax And enable the maximum configuration capacity of energy storage. E E,max exist Inside; (3) Step 5.3: With the goal of minimizing cost, construct the objective function of the energy storage configuration model regarding the uncertainty of load and new energy output using equation (4): (4) In equation (4): C total Total configuration cost; C ch The total cost of energy storage is given by equation (5); C sys The system operating cost is given by equation (6); (5) (6) In equations (5) and (6): C C Indicates the lifespan cost of energy storage; C inv For energy storage construction costs; C ope For energy storage operation and maintenance costs; The discount rate; For energy storage lifecycle; C E Cost of energy storage battery capacity; C P Cost per unit power; E E P represents the total configured capacity of energy storage. E C represents the total configured power of energy storage. o Daily operating and maintenance cost per unit; P c The daily charge / discharge power of the energy storage is T; T is the prediction period. Operating costs of thermal power units; , Costs associated with the start-up and shutdown of thermal power units; Cost per unit load shedding; To predict the system load shedding at time t within the prediction period; The cost of energy curtailment for new energy units; For the prediction period t Constantly discarding new energy sources.

3. The energy storage configuration method based on source-load growth prediction combination according to claim 2, characterized in that, The total energy storage cost in the energy storage configuration model C ch = C C + C inv + C ope ; in, C C The energy storage lifetime cost is represented by equation (7); (7) In equation (7), N This represents the total number of energy storage cycles within the planning period. For energy storage n The equivalent cycle is obtained from equation (8). c DOD The cost of performing a deep charge-discharge cycle on the battery is obtained from equation (9): (8) (9) In equations (8) and (9): N eq Equivalent coefficient; N en The number of battery cycles at 100% depth of discharge; c DOD Cost per cycle at 100% depth of discharge; D DOD This represents the actual depth of cycle discharge, and D DOD = SOC 1- SOC 2, SOC 1 indicates the initial charge or discharge capacity of the energy storage battery. SOC 2 indicates the final charge / discharge level of the energy storage battery; N roll This refers to the number of battery cycles at 100% depth of discharge.

4. The energy storage configuration method based on source-load growth prediction combination according to claim 2, characterized in that, The constraints of the energy storage planning model are constructed using equations (11)-(19): (11) (12) (13) (14) (15) (16) (17) (18) (19) In equations (11)-(19): P New,t , P C,t , P L,t Forecast t Real-time system of new energy output, energy storage output, and load. It is in energy storage working state, and ,in X in In energy storage charging state, X out It is in the energy storage and discharge state; P ij,t For prediction t Time Node i , j Power between, b ij For nodes i , j Internal susceptance θ i,t , θ j,t Forecast t Time Node i , j The phase angle; SOC t For prediction t At any given moment, the state of charge SOC min , SOC max These are the maximum and minimum permissible depth of discharge for the battery, respectively. E max This represents the maximum permissible capacity of the energy storage system.

5. An electronic device, comprising a memory and a processor, characterized in that, The memory is used to store a program that supports the processor in executing any of the energy storage configuration methods of claims 1-4, and the processor is configured to execute the program stored in the memory.

6. A computer-readable storage medium storing a computer program thereon, characterized in that, The computer program, when run by the processor, performs the steps of any of the energy storage configuration methods described in claims 1-4.

Images